NeoXam Agents — Platform
The governed agentic platform for investment operations.
Declare agents in a Git-governed catalog, gate them with evaluation baselines, run them through one policy-controlled runtime, and trace every step for audit — inside the platforms that already run your investment operations.
V0 runtime delivered · conversational mode & approvals in Beta · GA Q3 2026
How it works
Declare. Govern. Run. Observe.
Four verbs, one platform. Every agent follows the same governed path — from a versioned descriptor to a fully traced run in production.
Declare
Every agent is a versioned YAML descriptor in the Git catalog: model, prompt, tools, skills, typed input and output schemas, budgets. Nothing undeclared can run.
Govern
Changes ship through pull-request review. Evaluation baselines gate quality on every change, and per-tool allow/ask/deny policies control what an agent may do.
Run
Products invoke agents through one stable Runner API: typed input in, schema-validated output back, live progress streamed, runs cancellable, budgets enforced.
Observe
Every model and tool call emits an OpenTelemetry span with tokens, cost and duration — and lands in an append-only audit log designed for 7-year retention.
Capability map
What the platform does — and how mature each part is
We publish the status of every capability and never present roadmap as product. Shipped means running today; Beta is in build; GA lands in Q3 2026.
- Agents & catalogShippedDeclared, versioned agents in a Git-governed catalog — task mode running today, conversational mode in Beta.
- Governance & evalsShippedPR review on every change, evaluation baselines that can only rise, and allow/ask/deny policies on every tool call.
- Conversational & human-in-the-loopBetaMulti-turn assistants embedded in product screens, and approval cards for sensitive actions — in Beta build now.
- Security & trustShippedPer-agent RBAC, tenant isolation enforced in the database, scoped credential vaults and an append-only audit log.
- Observability & costShippedOpenTelemetry traces to Langfuse, hard per-run budgets, and cost dashboards per run, agent and tenant.
- Integrations & MCPShippedEvery tool is an MCP server; open standards throughout. Live connectors to NeoXam products land in Beta.
- Deployment & residencyShippedSaaS in the EU (Frankfurt) today; hybrid, on-premise and sovereign options arrive at GA in Q3 2026.
- ArchitectureRunner, harness, control plane versus execution plane — the technical view for CTOs and CISOs.
Under the hood
One runtime, two planes
Products call one stable Runner API. The control plane decides what may run; the execution plane runs it and accounts for every step.
Control plane
The Git catalog, tool policies, eval gates and the Console govern the fleet. Nothing undeclared is callable, injectable or provisionable.
Execution plane
One Runner authenticates every call, resolves the tenant, compiles the prompt, enforces budgets and validates every output against its schema.
Telemetry & audit
Every step emits an OpenTelemetry span to Langfuse; every run lands in an append-only audit log with the exact catalog version that executed.
NeoXam products
ARO · DataHub · GP · Impress · PMS — agents invoked in-screen
Runner API
auth · tenancy · rate limits · schema validation
Harness + MCP tools
agentic loop · model gateway (Claude today) · governed tool servers
Observability & audit
OpenTelemetry → Langfuse · append-only audit · cost per run
General availability comes in Q3 2026. The Early Adopter Program is open now.
A limited cohort, a one-year platform trial and three workshop streams — Business ROI, Compliance, Operational fit. Bring one workflow; leave with a governed agent and the evidence to certify it.